Warfarin Pharmacogenomics: Current Best Evidence

Journal of Thrombosis and Haemostasis, 13 (Suppl. 1): S266–S271 DOI: 10.1111/jth.12978

INVITED REVIEW

Warfarin pharmacogenomics: current best evidence

S. E. KIMMEL Department of Medicine and Department of Biostatistics and Epidemiology, Perelman University of Pennsylvania School of Medicine, Philadelphia, PA, USA

To cite this article: Kimmel SE. Warfarin pharmacogenomics: current best evidence. J Thromb Haemost 2015; 13 (Suppl. 1): S266–S71. data on genetic-based prescribing, highlight the impor- Summary. The utility of using genetic information to tance and distinction between clinical validity and clinical guide warfarin dosing has remained unclear based on utility, review the most recent evidence for clinical utility prior observational studies and small clinical trials. Two of warfarin pharmacogenetics, and consider possible appro- larger trials of warfarin and one of the acenocoumarol aches to using genetic information in warfarin-treated and phenprocoumon have recently been published. The patients. COAG trial addressed the incremental benefit of adding genetic information to clinical information and demon- Evaluating genetic-based medication dosing: when is it strated no benefit from the pharmacogenetic-based dosing ready for clinical use? strategy on the primary outcome. The EU-PACT UK trial compared an algorithm approach using genetic and There are several pieces of evidence that are important to clinical information to one that used a relatively fixed consider before using genetic information for medication starting dose. The pharmacogenetic-based algorithms dosing. These can be categorized as: Analytic Validity; improved the primary outcome. The study of acenocou- Clinical Validity; Clinical Utility; and Ethical, Legal, and marol and phenprocoumon compared a pharmacogenetic Social Issues (ELSI) [5,6]. with a clinical algorithm and demonstrated no benefit on Analytic Validity is how accurately and reliably the test the primary outcome. The evidence to date does not sup- measures the genotype of interest; that is, does the assay port an incremental benefit of adding genetic information provide accurate information about the genetic variants to clinical information on anticoagulation control. How- being used? ever, compared with fixed dosing, a pharmacogenetic Clinical Validity refers to how consistently and strongly algorithm can improve anticoagulation control. the genetic variants relate to the outcome of interest. Typ- ically this is determined from non-randomized studies Keywords: drug therapy; genetics; pharmacogenetics; examining the association between genetic variants and randomized controlled trials as topic; warfarin. outcomes. Clinical Utility refers to how likely the test is to signifi- cantly improve patient outcomes; that is, if one were to Introduction use genetic information to guide medication dosing, does it make a difference? Whether or not to use genetic information to guide and/ ELSI are any of these important issues that can arise or alter therapy remains a controversial issue in many from the use of genetic information in practice. fields [1–4]. The utility of using genetic information to Using warfarin as an example, analytic validity would guide warfarin dosing is one of the more debated. A large measure how accurately and reliably one can measure the body of literature has documented the association variants of interest in the laboratory in actual practice. If between genetic polymorphisms and warfarin mainte- the genotyping error rate is large, then the assay does not nance dose requirements, but such data do not answer have analytic validity and, of course, should not be used. the question: If one uses genetic information to select the Clinical validity would be demonstrated by high-quality, initial dosing of warfarin, will outcomes improve? This replicated studies demonstrating that patients with differ- review will discuss an approach to evaluating scientific ent genotypes require different warfarin doses to reach a therapeutic and stable level of anticoagulation. However, Correspondence: Stephen E. Kimmel, Department of Medicine and even if analytic and clinical validity are demonstrated, it Department of Biostatistics and Epidemiology, Perelman University of Pennsylvania School of Medicine, 923 Blockley Hall, 423 Guard- may still not be clear whether using genetic information ian Drive, Philadelphia, PA 19035, USA. at warfarin initiation will actually improve outcomes. EL- Tel:+1 215 898 1740; fax: +1 215 573 3106. SI may include whether patients are properly informed of E-mail: [email protected] their genotype and how it may affect not only their war-

© 2015 International Society on Thrombosis and Haemostasis Warfarin pharmacogenomics S267 farin therapy but other therapies as well, whether the farin based on the INR response to the drug and thus risk-benefit of genotype is adequately disclosed to correct for doses that may not be appropriate for an indi- patients, etc. vidual patient. These other considerations highlight the importance of testing the clinical utility of using genetics to guide warfarin dosing. Warfarin pharmacogenetics current evidence There is no one way to demonstrate clinical utility of Using the approach outlined above, one can assess the warfarin pharmacogenetics. However, it is clear that the current evidence for warfarin and thus make an informed randomized controlled trial (RCT) remains the gold stan- decision about whether genetic testing should be per- dard for comparing genetic-based strategies with other formed for patients treated with the drug. approaches. Thus, although observational studies have There are several genotyping assays for warfarin that suggested potential utility from genetic-based warfarin have demonstrated outstanding accuracy for the variants dosing [9,10], these studies are subjected to uncontrolled of interest (in particular, variants in the cytochrome P- confounding and epidemiological biases. 450 family 2 subfamily C polypeptide 9 enzyme RCTs can also be designed to test different hypotheses (CYP2C9) gene and the vitamin K epoxide reductase about warfarin pharmacogenetics. As there are many complex 1 (VKORC1)) and that have been FDA non-genetic factors that can alter warfarin response that approved for clinical use. (CYP2C9 is the major metabo- are commonly considered in practice, one might want to lizing enzyme for the S-enantiomer of warfarin, the more test whether the use of genetic information is useful above active form of the drug, and VKORC1 encodes the and beyond the clinical information readily available enzyme that is directly inhibited by warfarin.) Thus, ana- prior to starting warfarin (e.g., age, body size, interacting lytic validity has been clearly met for warfarin pharmaco- medications). Such a study would be designed so that the genetics. two study arms differ from each other only in the use of There is also a very large body of scientific data that genetic information, with one arm using this information demonstrate a consistent and strong relationship between and the other not using this information (but both arms certain genetic variants and warfarin dose requirements using clinical information). Another approach would be [7]. Variants in CYP2C9 and VKORC1 have been the to compare the use of both clinical and genetic informa- best studied and demonstrate consistent relationships with tion to guide warfarin dosing vs. using neither. Such a warfarin dose requirements. There are, however, some study would test whether a pharmacogenetic algorithm differences among racial groups. In particular, CYP2C9*2 (i.e., on that includes both genetic and non-genetic infor- and CYP2C9*3 are not as useful in African Americans, mation) is superior to an approach that used neither clini- most likely because of the low prevalence of these vari- cal nor genetic information (sometimes referred to as ants. Other variants in CYP2C9 that are more common ‘fixed-dose’ strategies). in African Americans appear to be more useful in predict- RCTs can also test efficacy or effectiveness. An efficacy ing warfarin dosing requirements [8], but to date there trial would test whether the use of genetic information are no FDA-approved assays for clinical use of these improves outcomes when used under carefully controlled variants. clinical care. For example, an efficacy trial would try to It is important to note that the studies that have estab- ensure that the dose titration of warfarin after the initial lished a relationship between genetic variants and warfa- dosing period (the period in which genetics is used to rin dose requirements have almost all been done by select the dose) was uniform between study arms. It might relating patients’ stable maintenance dose to their genetic also blind both patients and clinicians to both the variants. The analyses assess the relationship between the patients’ genotype and their warfarin dose to prevent dif- genetic variants and the maintenance dose. These studies ferences in care between the two arms that are related to are limited in that only patients reaching a stable dose knowing which arm a patient was in. In contrast, effec- are included; those that may be most affected by altered tiveness trials would test whether the use of genetic infor- pharmacokinetics or pharmacodynamics of warfarin – mation improves outcomes under usual clinical practice those that fail to achieve a stable dose – are typically settings. Such a trial could allow for differences in dose excluded. In addition, just because a genetic variant is titration between study arms and allow patients and clini- associated with the ultimate warfarin maintenance dose, it cians to know their warfarin dose. As a more extreme does not mean that starting patients at their genetically example, an effectiveness trial could simply randomize predicted dose from the start of therapy will improve patients and physicians to knowing or not knowing a their anticoagulation control, time to reach a stable main- patient’s genotype and allow the clinicians to use the tenance dose, or risk of adverse events. In addition, there information in any way they see fit. It can be argued that are many other patient and environmental factors that without efficacy there can be no effectiveness, but this can influence warfarin response, including age, body size, may not always be the case for pharmacogenetics. For interacting medications, comorbidities, and medication example, providing genetic information to clinicians and adherence. Further, clinicians can titrate the dose of war- patients prior to warfarin therapy could alter both patient

© 2015 International Society on Thrombosis and Haemostasis S268 S. E. Kimmel and physician behavior (e.g., monitoring frequency, single nucleotide polymorphisms (SNPs): CYP2C9*2, patient adherence) in way that has positive or negative CYP2C9*3, and a single VKORC1 SNP (rs9923231). effects on outcomes. Prior to 2013, there were several small RCTs of warfa- Clarification of optimal anticoagulation through genetics rin pharmacogenetics [11–14]. One trial of 206 patients compared a pharmacogenetic algorithm that included The Clarification of Optimal Anticoagulation through both genetic and clinical information with standard Genetics (COAG) trial was the largest trial to date, ran- (‘fixed’) dosing [11]. This study did not demonstrate a ben- domizing 1015 patients and making it larger than the efit on the primary outcome of percent out-of-range INRs, other two trials combined [15]. The trial used a pharma- although it did suggest that those with either zero or more cogenetic dosing algorithm that included clinical and than one variants in the two genes studied (CYP2C9 and genetic factors. The study did not use a loading dose VKORC1) might benefit from the pharmacogenetic algo- strategy although the first dose was calculated without rithm. Another trial of 230 patients compared a pharma- considering the effects of CYP2C9 because CYP2C9 vari- cogenetic algorithm that included both genetic and clinical ants are known to have little influence on INR response information with a clinical-only algorithm that used clini- early in therapy and because of the concern that decreas- cal information to inform the starting dose [13]. This ing the first dose in slow metabolizers would delay the study also did not demonstrate a benefit on time in thera- time until the INR is therapeutic. The comparison group peutic range with pharmacogenetic dosing. Of note, both was a clinical algorithm that used all of the same clinical of the latter studies showed that the pharmacogenetic factors as in the pharmacogenetic arm but did not use algorithms clearly predicted maintenance dose better than genetic information. The study included dosing algo- the comparison arms. Two other studies used only genetic rithms for the first 3 days of therapy and dose-revision information to determine dose and compared it with a algorithms on days 4 and 5 for both study arms. The starting dose that did not include either genetic or clinical COAG trial therefore tested whether genetics provides information [12,14]. In one of the studies, among 101 incremental benefit, above and beyond what can be deter- patients, the genetic-based dosing arm had a shorter time mined from clinical information alone. The trial also to achieving warfarin maintenance dose [14]. The other performed dose titration after the initial dosing phase study of 191 patients demonstrated better anticoagulation (the intervention phase) using a computer-based algo- control in the genetic-guided arm and less minor bleeding, rithm that was applied equally across both arms and but suffered from high dropout rates [12]. Taken together, across all sites. these studies did not provide definitive evidence for a ben- The study was unique in that it was the only trial that efit of pharmacogenetic-based warfarin dosing. blinded patients and clinicians to the dose of warfarin In 2013, three independent clinical trials were published during the primary outcome period. As noted in the study simultaneously in the New England Journal of Medicine protocol (see supplement in [15]): ‘The primary outcome [15–17]. These studies were larger than prior studies. They of the study (PTTR) could therefore be affected not only also had different designs that addressed different ques- by warfarin dose or warfarin dose adjustments, but by tions. Several characteristics of these studies are illus- other, post-randomization factors that could differ if trated in the Table 1. All studies used the same three study arms and drug dose are not blinded. These include differential dropout, protocol deviations, cross-overs (e.g., genotyping those in the non-genotype-guided arm), differences in adherence, and differences in patient care. These Table 1 Comparison of recent RCTs* would be particularly problematic if the occurrence of COAG EU-PACT UK EU-PACT N/G these post-randomization factors both differed by study arm and were also related to anticoagulation control, as Drug Warfarin Warfarin Acenocoumarol/ might be expected. For example, patients who know or Phenprocoumon PGx arm Algorithm Algorithm Algorithm suspect that they are in the clinical arm who are having (w/loading) difficulty with anticoagulation early in therapy (those Comparison Clinical Fixed dose Clinical who might contribute the most to any differences by arm algorithm (By Age) algorithm study arm) may be more likely to withdraw from the pro- Blinding Double Single Single tocol than those in the genotyping arm because clinicians African 27% 1.1% 0% Americans (or the participants themselves) would want to know their Primary PTTR at PTTR at PTTR at genotype or manage their dosing themselves. Another outcome 4 weeks 12 weeks 12 weeks potential bias is that patients on very low doses, who Primary Negative Positive Negative would be known or presumed to carry genetic variants results that make them sensitive to warfarin, may be managed *N/G, Netherlands/Greece; PGx, pharmacogenetic; PTTR, percent more carefully: they may be counselled more aggressively time in therapeutic range. about dietary adherence and they may be more

© 2015 International Society on Thrombosis and Haemostasis Warfarin pharmacogenomics S269 meticulous about avoiding interacting medications, be macogenetic-based dose-revision algorithm on days 4 and asked to come for more frequent INR monitoring beyond 5; this was the same algorithm as used in COAG. The that required by protocol, and/or seek such extra moni- comparison arm dosed all patients 75 years of age and toring. As another example, physicians could be more under the same dose (20 mg over the first 3 days) and likely to withdraw a patient from the study if they felt patients greater than 75 years of age the same dose (15 mg that one of the algorithm-based approaches was not over the first 3 days). There was no dose-revision algorithm working and wanted to dose patients themselves (e.g., in in this arm. The trial thus compared a strategy using clini- a patient taking too long to reach adequate levels of anti- cal and genetic information in a formal dosing algorithm coagulation, thus delaying hospital discharge). Patients with a strategy that used neither (except for reducing the may also be scheduled for extra study visits and be more dose by 5 mg on the first day of therapy in those over likely to be adherent with these visits (e.g., less missed vis- 75 years of age). After the initial 5 days of therapy, dose its) if they know they are ‘more sensitive’ to warfarin. All titration was performed as per local practice at each site, of these scenarios could bias the results in manners that although each site did use a formal dose titration method. are difficult if not impossible to measure and control.’ The EU-PACT UK trial tested a different hypothesis The COAG trial also included 27% African Americans. than the COAG trial. EU-PACT UK tested whether an Prior to COAG, the total number of African Americans approach that uses a formal dosing algorithm that incor- in similar warfarin pharmacogenetic trials was 3. The porates both genetic and clinical information to determine trial’s a priori hypothesis was that there would be differ- dose over the first 5 days of therapy was superior to an ences between African Americans and other racial groups approach that used neither (except for age in the first because it was known that dosing algorithms performed 3 days). less well in African Americans even though the algo- The EU-PACT UK study also did not blind patients or rithms included a variable to specify race [18]. As such, clinicians to the dosing strategy. Although patients and the trial stratified the randomization by race. clinicians did not know their genotype, they could easily The study demonstrated no benefit of pharmacogenet- discern which study arm they were in because the dosing ic-based dosing vs. clinical-based dosing on the primary regimens were different and because only the genotype outcome of percent time in therapeutic range (PTTR) of arm had genotyping done at the start of the trial. The the INR at 4 weeks: PTTR of 45.2% vs. 45.4%, respec- EU-PACT UK trial included only 1.1% African Ameri- tively. There was also no significant benefit of pharmaco- cans and thus could not determine the effects of their genetic-based dosing in the pre-specified subset of patients pharmacogenetic strategy in this group of patients. in whom the two algorithms predicted at least a The study enrolled a total of 455 patients and the pri- 1 mg dayÀ1 dose difference for warfarin. However, there mary outcome was PTTR at 3 months. The mean PTTR was a statistically significant difference between African was better in the pharmacogenetic group (67.4%) than in Americans and non-African Americans. African Ameri- the control group (60.3%; adjusted difference, 7.0 per- cans fared worse with the pharmacogenetic algorithm centage points; P < 0.001). Patients in the pharmacoge- than with the clinical algorithms (PTTR 35.2% vs. netic arm also were less likely to have an elevated INR 43.5%, respectively; adjusted mean difference, À8.3%; (4.0 or higher) and had a shorter time to the first thera- P = 0.01), while there was no statistically significant bene- peutic INR than patients in the control arm. There were fit of pharmacogenetic vs. clinical dosing in non-African no major bleeds in the trial. Americans (PTTR 48.8% vs. 46.1%, respectively; adjusted mean difference, 2.8%; P = 0.15). There was no EU-PACT Netherlands/Greece (N/G) difference in the principal secondary outcome of INR of 4 or more, major bleeding, or thromboembolism. The third trial did not use warfarin but rather the couma- rin anticoagulants acenocoumarol and phenprocoumon [16]. Originally designed as two separate trials of the two European pharmacogenetics of anticoagulant therapy UK different drugs, the data from the two trials were com- The European Pharmacogenetics of Anticoagulant Ther- bined for analyses because of low enrollment. Similar to apy (EU-PACT) UK study took a different approach the COAG trial, the EU-PACT N/G trial compared a from the COAG trial [17]. EU-PACT UK compared a formal pharmacogenetic algorithm that included clinical pharmacogenetic-based dosing strategy to a relatively and genetic factors with a clinical algorithm that used the fixed-dose strategy. The pharmacogenetic arm used clini- same clinical factors as the pharmacogenetic but did not cal and genetic data (using the same genetic variants as use genetic information. Similar to the EU-PACT UK COAG but different clinical factors and a different dos- trial, the study did not blind patients and clinicians to the ing equation) and also loaded patients up-front by giving dosing strategies. Also, management of dose titration was a higher proportion of the first 3 days of warfarin on day not required to be standardized between the study arms 1 than on day 2 and a higher proportion on day 2 than although each site used its own standardized dose titra- day 3. The pharmacogenetic algorithm also used a phar- tion algorithms. The trial had the same primary endpoint

© 2015 International Society on Thrombosis and Haemostasis S270 S. E. Kimmel as the EU-PACT UK trial (PTTR at 3 months) and design among the trials make combining data problematic enrolled no African Americans. and the most valid results are those that come for each The primary results of the trial demonstrated no benefit individual trial. to the pharmacogenetic approach. The PTTR was 61.6% in the pharmacogenetic arm and 60.2% in the clinical Disclosure of Conflict of Interests arm (P = 0.52). There also were no significant differences between the two groups for multiple secondary outcomes, The author reports personal fees from Pfizer, outside the including INRs ≥ 4.0, time to reach therapeutic INR, and submitted work. patients with stable dose within 12 weeks of starting therapy. There were no differences as well in bleeding or References thromboembolic outcomes. In an analysis of PTTR at different time points in the study, there was a significant 1 Ginsburg GS, Voora D. The long and winding road to warfarin 55 – difference in PTTR at 4 weeks between the pharmacoge- pharmacogenetic testing. J Am Coll Cardiol 2010; : 2813 5. 2 Burke W, Laberge AM, Press N. Debating clinical utility. Public netic arm (PTTR 52.5%) and the clinical arm (PTTR Health Genomics 2010; 13: 215–23. 47.5%, P = 0.02). 3 Ashley EA, Hershberger RE, Caleshu C, Ellinor PT, Garcia JG, Herrington DM, Ho CY, Johnson JA, Kittner SJ, Macrae CA, Mudd-Martin G, Rader DJ, Roden DM, Scholes D, Sellke FW, Conclusions Towbin JA, van Eyk J, Worrall BB. Genetics and cardiovascular disease: a policy statement from the American Heart Association. In summary, the current evidence does not support an Circulation 2012; 126: 142–57. incremental benefit on anticoagulation control of using 4 Woodcock J, Lesko LJ. Pharmacogenetics–tailoring treatment CYP2C9*2, CYP2C9*3, and VKORC1 rs9923231 to for the outliers. N Engl J Med 2009; 360: 811–3. determine the initial warfarin dosing above and beyond 5 Centers for Disease Control and Prevention (CDC). Genomic what is available from clinical information. The EU- Testing. http://www.cdc.gov/genomics/gtesting/ACCE/index.htm. Accessed 2 February 2015. PACT UK study demonstrated that a pharmacogenetic 6 Teutsch SM, Bradley LA, Palomaki GE, Haddow JE, Piper M, algorithm was superior to an approach that did not use a Calonge N, Dotson WD, Douglas MP, Berg AO. The Evaluation formal clinical algorithm and only incorporated age. The of Genomic Applications in Practice and Prevention (EGAPP) COAG study demonstrated worse anticoagulation control Initiative: methods of the EGAPP Working Group. Genet Med 11 – in African Americans with the use of a pharmacogenetic 2009; :3 14. 7 Johnson JA, Cavallari LH. Warfarin pharmacogenetics. Trends algorithm compared with a clinical algorithm. Although, Cardiovasc Med 2015; 25:33–41. taken together, the trials suggest that a clinical-only algo- 8 Suarez-Kurtz G, Botton MR. Pharmacogenomics of warfarin in rithm might be the preferable approach to dosing warfa- populations of African descent. Br J Clin Pharmacol 2013; 75: rin, no study to date has directly compared a clinical-only 334–46. algorithm with a fixed-dose approach. None of the studies 9 Epstein RS, Moyer TP, Aubert RE, O Kane DJ, Xia F, Verbrugge RR, Gage BF, Teagarden JR. Warfarin genotyping reduces hospi- were powered to examine major bleeding or thromboem- talization rates results from the MM-WES (Medco-Mayo Warfa- bolic events. In addition, these studies were carried out in rin Effectiveness study). J Am Coll Cardiol 2010; 55: 2804–12. carefully managed protocols for dose titration with fre- 10 Anderson JL, Horne BD, Stevens SM, Woller SC, Samuelson quent follow-up and do not necessarily reflect the effects KM, Mansfield JW, Robinson M, Barton S, Brunisholz K, of pharmacogenetic-based dosing in different practice set- Mower CP, Huntinghouse JA, Rollo JS, Siler D, Bair TL, Knight S, Muhlestein JB, Carlquist JF. A randomized and clini- tings. Pragmatic effectiveness trials would be needed to cal effectiveness trial comparing two pharmacogenetic algorithms address the effects of pharmacogenetic-based dosing in and standard care for individualizing warfarin dosing (Couma- broad-based practice and with less control over dose titra- Gen-II). Circulation 2012; 125: 1997–2005. tion methods. 11 Anderson JL, Horne BD, Stevens SM, Grove AS, Barton S, It should be noted that all of the recent trials demon- Nicholas ZP, Kahn SFS, May HT, Samuelson KM, Muhlestein JB, Carlquist JF; Couma-Gen Investigators. Randomized trial of strated clear superiority from the pharmacogenetic algo- genotype-guided versus standard warfarin dosing in patients ini- rithms in predicting the final warfarin maintenance dose. tiating oral anticoagulation. Circulation 2007; 116: 2563–70. The disconnect between predicting maintenance dose and 12 Caraco Y, Blotnick S, Muszkat M. CYP2C9 genotype-guided improving anticoagulation control in many studies illus- warfarin prescribing enhances the efficacy and safety of anticoag- trates the important distinction between clinical validity ulation: a prospective randomized controlled study. Clin Pharma- col Ther 2008; 83: 460–70. (strong relationship between genetic variants and dose 13 Burmester JK, Berg RL, Yale SH, Rottscheit CM, Glurich IE, requirements) and clinical utility (the use of genetics to Schmelzer JR, Caldwell MD. A randomized controlled trial of alter dose in practice). genotype-based Coumadin initiation. Genet Med 2011; 13: 509– An additional study, the Genetics informatics trial 18. (GIFT) is underway to examine the effects of pharmaco- 14 Wang M, Lang X, Cui S, Fei K, Zou L, Cao J, Wang L, Zhang S, Wu X, Wang Y, Ji Q. Clinical application of pharmacogenet- genetic dosing on outcomes in patients undergoing ortho- ic-based warfarin-dosing algorithm in patients of Han nationality pedic surgery [19]. Although meta-analyses are also being after rheumatic valve replacement: a randomized and controlled published of the existing trials, the differences in study trial. Int J Med Sci 2012; 9: 472–9.

15 Kimmel SE, French B, Kasner SE, Johnson JA, Anderson JL, Stafberg C, Zhang JE, Leathart JB, Kohnke H, Maitland-van Gage BF, Rosenberg YD, Eby CS, Madigan RA, McBane RB, der Zee AH, Williamson PR, Daly AK, Avery P, Kamali F, Abdel-Rahman SZ, Stevens SM, Yale S, Mohler ER III, Fang Wadelius M. A randomized trial of genotype-guided dosing of MC, Shah V, Horenstein RB, Limdi NA, Muldowney JA III, warfarin. N Engl J Med 2013;369:2294–303. Gujral J, et al. A pharmacogenetic versus a clinical algorithm for 18 Schelleman H, Chen J, Chen Z, Christie J, Newcomb CW, Bren- warfarin dosing. N Engl J Med 2013; 369: 2283–93. singer CM, Price M, Whitehead AS, Kealey C, Thorn CF, Sama- 16 Verhoef TI, Ragia G, de Boer A, Barallon R, Kolovou G, Kol- ha FF, Kimmel SE. Dosing algorithms to predict warfarin ovou V, Konstantinides S, Le Cessie S, Maltezos E, van der maintenance dose in Caucasians and African Americans. Clin Meer FJ, Redekop WK, Remkes M, Rosendaal FR, van Schie Pharmacol Ther 2008; 84: 332–9. RM, Tavridou A, Tziakas D, Wadelius M, Manolopoulos VG, 19 Do EJ, Lenzini P, Eby CS, Bass AR, McMillin GA, Stevens Maitland-van der Zee AH. A randomized trial of genotype- SM, Woller SC, Pendleton RC, Anderson JL, Proctor P, Nunley guided dosing of acenocoumarol and phenprocoumon. N Engl J RM, Davila-Roman V, Gage BF. Genetics informatics trial Med 2013;369:2304–12. (GIFT) of warfarin to prevent deep vein thrombosis (DVT): 17 Pirmohamed M, Burnside G, Eriksson N, Jorgensen AL, Toh rationale and study design. Pharmacogenomics J 2012; 12: CH, Nicholson T, Kesteven P, Christersson C, Wahlstrom B, 417–24.